2 * Read-Copy Update mechanism for mutual exclusion
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright IBM Corporation, 2008
20 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
21 * Manfred Spraul <manfred@colorfullife.com>
22 * Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version
24 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
25 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
27 * For detailed explanation of Read-Copy Update mechanism see -
30 #include <linux/types.h>
31 #include <linux/kernel.h>
32 #include <linux/init.h>
33 #include <linux/spinlock.h>
34 #include <linux/smp.h>
35 #include <linux/rcupdate.h>
36 #include <linux/interrupt.h>
37 #include <linux/sched.h>
38 #include <linux/nmi.h>
39 #include <asm/atomic.h>
40 #include <linux/bitops.h>
41 #include <linux/module.h>
42 #include <linux/completion.h>
43 #include <linux/moduleparam.h>
44 #include <linux/percpu.h>
45 #include <linux/notifier.h>
46 #include <linux/cpu.h>
47 #include <linux/mutex.h>
48 #include <linux/time.h>
52 /* Data structures. */
54 #define RCU_STATE_INITIALIZER(name) { \
55 .level = { &name.node[0] }, \
57 NUM_RCU_LVL_0, /* root of hierarchy. */ \
60 NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \
62 .signaled = RCU_GP_IDLE, \
65 .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
66 .orphan_cbs_list = NULL, \
67 .orphan_cbs_tail = &name.orphan_cbs_list, \
69 .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
71 .n_force_qs_ngp = 0, \
74 struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state);
75 DEFINE_PER_CPU(struct rcu_data, rcu_sched_data);
77 struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
78 DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
82 * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s
83 * permit this function to be invoked without holding the root rcu_node
84 * structure's ->lock, but of course results can be subject to change.
86 static int rcu_gp_in_progress(struct rcu_state *rsp)
88 return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum);
92 * Note a quiescent state. Because we do not need to know
93 * how many quiescent states passed, just if there was at least
94 * one since the start of the grace period, this just sets a flag.
96 void rcu_sched_qs(int cpu)
100 rdp = &per_cpu(rcu_sched_data, cpu);
101 rdp->passed_quiesc_completed = rdp->completed;
103 rdp->passed_quiesc = 1;
104 rcu_preempt_note_context_switch(cpu);
107 void rcu_bh_qs(int cpu)
109 struct rcu_data *rdp;
111 rdp = &per_cpu(rcu_bh_data, cpu);
112 rdp->passed_quiesc_completed = rdp->completed;
114 rdp->passed_quiesc = 1;
118 DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
119 .dynticks_nesting = 1,
122 #endif /* #ifdef CONFIG_NO_HZ */
124 static int blimit = 10; /* Maximum callbacks per softirq. */
125 static int qhimark = 10000; /* If this many pending, ignore blimit. */
126 static int qlowmark = 100; /* Once only this many pending, use blimit. */
128 module_param(blimit, int, 0);
129 module_param(qhimark, int, 0);
130 module_param(qlowmark, int, 0);
132 static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
133 static int rcu_pending(int cpu);
136 * Return the number of RCU-sched batches processed thus far for debug & stats.
138 long rcu_batches_completed_sched(void)
140 return rcu_sched_state.completed;
142 EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
145 * Return the number of RCU BH batches processed thus far for debug & stats.
147 long rcu_batches_completed_bh(void)
149 return rcu_bh_state.completed;
151 EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
154 * Does the CPU have callbacks ready to be invoked?
157 cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
159 return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL];
163 * Does the current CPU require a yet-as-unscheduled grace period?
166 cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
168 return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp);
172 * Return the root node of the specified rcu_state structure.
174 static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
176 return &rsp->node[0];
180 * Record the specified "completed" value, which is later used to validate
181 * dynticks counter manipulations and CPU-offline checks. Specify
182 * "rsp->completed - 1" to unconditionally invalidate any future dynticks
183 * manipulations and CPU-offline checks. Such invalidation is useful at
184 * the beginning of a grace period.
186 static void dyntick_record_completed(struct rcu_state *rsp, long comp)
188 rsp->dynticks_completed = comp;
194 * Recall the previously recorded value of the completion for dynticks.
196 static long dyntick_recall_completed(struct rcu_state *rsp)
198 return rsp->dynticks_completed;
202 * If the specified CPU is offline, tell the caller that it is in
203 * a quiescent state. Otherwise, whack it with a reschedule IPI.
204 * Grace periods can end up waiting on an offline CPU when that
205 * CPU is in the process of coming online -- it will be added to the
206 * rcu_node bitmasks before it actually makes it online. The same thing
207 * can happen while a CPU is in the process of coming online. Because this
208 * race is quite rare, we check for it after detecting that the grace
209 * period has been delayed rather than checking each and every CPU
210 * each and every time we start a new grace period.
212 static int rcu_implicit_offline_qs(struct rcu_data *rdp)
215 * If the CPU is offline, it is in a quiescent state. We can
216 * trust its state not to change because interrupts are disabled.
218 if (cpu_is_offline(rdp->cpu)) {
223 /* If preemptable RCU, no point in sending reschedule IPI. */
224 if (rdp->preemptable)
227 /* The CPU is online, so send it a reschedule IPI. */
228 if (rdp->cpu != smp_processor_id())
229 smp_send_reschedule(rdp->cpu);
236 #endif /* #ifdef CONFIG_SMP */
241 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
243 * Enter nohz mode, in other words, -leave- the mode in which RCU
244 * read-side critical sections can occur. (Though RCU read-side
245 * critical sections can occur in irq handlers in nohz mode, a possibility
246 * handled by rcu_irq_enter() and rcu_irq_exit()).
248 void rcu_enter_nohz(void)
251 struct rcu_dynticks *rdtp;
253 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
254 local_irq_save(flags);
255 rdtp = &__get_cpu_var(rcu_dynticks);
257 rdtp->dynticks_nesting--;
258 WARN_ON_ONCE(rdtp->dynticks & 0x1);
259 local_irq_restore(flags);
263 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
265 * Exit nohz mode, in other words, -enter- the mode in which RCU
266 * read-side critical sections normally occur.
268 void rcu_exit_nohz(void)
271 struct rcu_dynticks *rdtp;
273 local_irq_save(flags);
274 rdtp = &__get_cpu_var(rcu_dynticks);
276 rdtp->dynticks_nesting++;
277 WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
278 local_irq_restore(flags);
279 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
283 * rcu_nmi_enter - inform RCU of entry to NMI context
285 * If the CPU was idle with dynamic ticks active, and there is no
286 * irq handler running, this updates rdtp->dynticks_nmi to let the
287 * RCU grace-period handling know that the CPU is active.
289 void rcu_nmi_enter(void)
291 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
293 if (rdtp->dynticks & 0x1)
295 rdtp->dynticks_nmi++;
296 WARN_ON_ONCE(!(rdtp->dynticks_nmi & 0x1));
297 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
301 * rcu_nmi_exit - inform RCU of exit from NMI context
303 * If the CPU was idle with dynamic ticks active, and there is no
304 * irq handler running, this updates rdtp->dynticks_nmi to let the
305 * RCU grace-period handling know that the CPU is no longer active.
307 void rcu_nmi_exit(void)
309 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
311 if (rdtp->dynticks & 0x1)
313 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
314 rdtp->dynticks_nmi++;
315 WARN_ON_ONCE(rdtp->dynticks_nmi & 0x1);
319 * rcu_irq_enter - inform RCU of entry to hard irq context
321 * If the CPU was idle with dynamic ticks active, this updates the
322 * rdtp->dynticks to let the RCU handling know that the CPU is active.
324 void rcu_irq_enter(void)
326 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
328 if (rdtp->dynticks_nesting++)
331 WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
332 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
336 * rcu_irq_exit - inform RCU of exit from hard irq context
338 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
339 * to put let the RCU handling be aware that the CPU is going back to idle
342 void rcu_irq_exit(void)
344 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
346 if (--rdtp->dynticks_nesting)
348 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
350 WARN_ON_ONCE(rdtp->dynticks & 0x1);
352 /* If the interrupt queued a callback, get out of dyntick mode. */
353 if (__get_cpu_var(rcu_sched_data).nxtlist ||
354 __get_cpu_var(rcu_bh_data).nxtlist)
361 * Snapshot the specified CPU's dynticks counter so that we can later
362 * credit them with an implicit quiescent state. Return 1 if this CPU
363 * is in dynticks idle mode, which is an extended quiescent state.
365 static int dyntick_save_progress_counter(struct rcu_data *rdp)
371 snap = rdp->dynticks->dynticks;
372 snap_nmi = rdp->dynticks->dynticks_nmi;
373 smp_mb(); /* Order sampling of snap with end of grace period. */
374 rdp->dynticks_snap = snap;
375 rdp->dynticks_nmi_snap = snap_nmi;
376 ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0);
383 * Return true if the specified CPU has passed through a quiescent
384 * state by virtue of being in or having passed through an dynticks
385 * idle state since the last call to dyntick_save_progress_counter()
388 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
395 curr = rdp->dynticks->dynticks;
396 snap = rdp->dynticks_snap;
397 curr_nmi = rdp->dynticks->dynticks_nmi;
398 snap_nmi = rdp->dynticks_nmi_snap;
399 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
402 * If the CPU passed through or entered a dynticks idle phase with
403 * no active irq/NMI handlers, then we can safely pretend that the CPU
404 * already acknowledged the request to pass through a quiescent
405 * state. Either way, that CPU cannot possibly be in an RCU
406 * read-side critical section that started before the beginning
407 * of the current RCU grace period.
409 if ((curr != snap || (curr & 0x1) == 0) &&
410 (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) {
415 /* Go check for the CPU being offline. */
416 return rcu_implicit_offline_qs(rdp);
419 #endif /* #ifdef CONFIG_SMP */
421 #else /* #ifdef CONFIG_NO_HZ */
425 static int dyntick_save_progress_counter(struct rcu_data *rdp)
430 static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
432 return rcu_implicit_offline_qs(rdp);
435 #endif /* #ifdef CONFIG_SMP */
437 #endif /* #else #ifdef CONFIG_NO_HZ */
439 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
441 static void record_gp_stall_check_time(struct rcu_state *rsp)
443 rsp->gp_start = jiffies;
444 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
447 static void print_other_cpu_stall(struct rcu_state *rsp)
452 struct rcu_node *rnp = rcu_get_root(rsp);
454 /* Only let one CPU complain about others per time interval. */
456 spin_lock_irqsave(&rnp->lock, flags);
457 delta = jiffies - rsp->jiffies_stall;
458 if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
459 spin_unlock_irqrestore(&rnp->lock, flags);
462 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
465 * Now rat on any tasks that got kicked up to the root rcu_node
466 * due to CPU offlining.
468 rcu_print_task_stall(rnp);
469 spin_unlock_irqrestore(&rnp->lock, flags);
471 /* OK, time to rat on our buddy... */
473 printk(KERN_ERR "INFO: RCU detected CPU stalls:");
474 rcu_for_each_leaf_node(rsp, rnp) {
475 rcu_print_task_stall(rnp);
476 if (rnp->qsmask == 0)
478 for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
479 if (rnp->qsmask & (1UL << cpu))
480 printk(" %d", rnp->grplo + cpu);
482 printk(" (detected by %d, t=%ld jiffies)\n",
483 smp_processor_id(), (long)(jiffies - rsp->gp_start));
484 trigger_all_cpu_backtrace();
486 force_quiescent_state(rsp, 0); /* Kick them all. */
489 static void print_cpu_stall(struct rcu_state *rsp)
492 struct rcu_node *rnp = rcu_get_root(rsp);
494 printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
495 smp_processor_id(), jiffies - rsp->gp_start);
496 trigger_all_cpu_backtrace();
498 spin_lock_irqsave(&rnp->lock, flags);
499 if ((long)(jiffies - rsp->jiffies_stall) >= 0)
501 jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
502 spin_unlock_irqrestore(&rnp->lock, flags);
504 set_need_resched(); /* kick ourselves to get things going. */
507 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
510 struct rcu_node *rnp;
512 delta = jiffies - rsp->jiffies_stall;
514 if ((rnp->qsmask & rdp->grpmask) && delta >= 0) {
516 /* We haven't checked in, so go dump stack. */
517 print_cpu_stall(rsp);
519 } else if (rcu_gp_in_progress(rsp) && delta >= RCU_STALL_RAT_DELAY) {
521 /* They had two time units to dump stack, so complain. */
522 print_other_cpu_stall(rsp);
526 #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
528 static void record_gp_stall_check_time(struct rcu_state *rsp)
532 static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
536 #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
539 * Update CPU-local rcu_data state to record the newly noticed grace period.
540 * This is used both when we started the grace period and when we notice
541 * that someone else started the grace period.
543 static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
546 rdp->passed_quiesc = 0;
547 rdp->gpnum = rsp->gpnum;
551 * Did someone else start a new RCU grace period start since we last
552 * checked? Update local state appropriately if so. Must be called
553 * on the CPU corresponding to rdp.
556 check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
561 local_irq_save(flags);
562 if (rdp->gpnum != rsp->gpnum) {
563 note_new_gpnum(rsp, rdp);
566 local_irq_restore(flags);
571 * Start a new RCU grace period if warranted, re-initializing the hierarchy
572 * in preparation for detecting the next grace period. The caller must hold
573 * the root node's ->lock, which is released before return. Hard irqs must
577 rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
578 __releases(rcu_get_root(rsp)->lock)
580 struct rcu_data *rdp = rsp->rda[smp_processor_id()];
581 struct rcu_node *rnp = rcu_get_root(rsp);
583 if (!cpu_needs_another_gp(rsp, rdp)) {
584 spin_unlock_irqrestore(&rnp->lock, flags);
588 /* Advance to a new grace period and initialize state. */
590 WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT);
591 rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
592 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
593 record_gp_stall_check_time(rsp);
594 dyntick_record_completed(rsp, rsp->completed - 1);
595 note_new_gpnum(rsp, rdp);
598 * Because this CPU just now started the new grace period, we know
599 * that all of its callbacks will be covered by this upcoming grace
600 * period, even the ones that were registered arbitrarily recently.
601 * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL.
603 * Other CPUs cannot be sure exactly when the grace period started.
604 * Therefore, their recently registered callbacks must pass through
605 * an additional RCU_NEXT_READY stage, so that they will be handled
606 * by the next RCU grace period.
608 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
609 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
611 /* Special-case the common single-level case. */
612 if (NUM_RCU_NODES == 1) {
613 rcu_preempt_check_blocked_tasks(rnp);
614 rnp->qsmask = rnp->qsmaskinit;
615 rnp->gpnum = rsp->gpnum;
616 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */
617 spin_unlock_irqrestore(&rnp->lock, flags);
621 spin_unlock(&rnp->lock); /* leave irqs disabled. */
624 /* Exclude any concurrent CPU-hotplug operations. */
625 spin_lock(&rsp->onofflock); /* irqs already disabled. */
628 * Set the quiescent-state-needed bits in all the rcu_node
629 * structures for all currently online CPUs in breadth-first
630 * order, starting from the root rcu_node structure. This
631 * operation relies on the layout of the hierarchy within the
632 * rsp->node[] array. Note that other CPUs will access only
633 * the leaves of the hierarchy, which still indicate that no
634 * grace period is in progress, at least until the corresponding
635 * leaf node has been initialized. In addition, we have excluded
636 * CPU-hotplug operations.
638 * Note that the grace period cannot complete until we finish
639 * the initialization process, as there will be at least one
640 * qsmask bit set in the root node until that time, namely the
641 * one corresponding to this CPU, due to the fact that we have
644 rcu_for_each_node_breadth_first(rsp, rnp) {
645 spin_lock(&rnp->lock); /* irqs already disabled. */
646 rcu_preempt_check_blocked_tasks(rnp);
647 rnp->qsmask = rnp->qsmaskinit;
648 rnp->gpnum = rsp->gpnum;
649 spin_unlock(&rnp->lock); /* irqs remain disabled. */
652 rnp = rcu_get_root(rsp);
653 spin_lock(&rnp->lock); /* irqs already disabled. */
654 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
655 spin_unlock(&rnp->lock); /* irqs remain disabled. */
656 spin_unlock_irqrestore(&rsp->onofflock, flags);
660 * Advance this CPU's callbacks, but only if the current grace period
661 * has ended. This may be called only from the CPU to whom the rdp
665 rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
670 local_irq_save(flags);
671 completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */
673 /* Did another grace period end? */
674 if (rdp->completed != completed_snap) {
676 /* Advance callbacks. No harm if list empty. */
677 rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
678 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
679 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
681 /* Remember that we saw this grace-period completion. */
682 rdp->completed = completed_snap;
684 local_irq_restore(flags);
688 * Clean up after the prior grace period and let rcu_start_gp() start up
689 * the next grace period if one is needed. Note that the caller must
690 * hold rnp->lock, as required by rcu_start_gp(), which will release it.
692 static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags)
693 __releases(rcu_get_root(rsp)->lock)
695 WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
696 rsp->completed = rsp->gpnum;
697 rsp->signaled = RCU_GP_IDLE;
698 rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
699 rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */
703 * Similar to cpu_quiet(), for which it is a helper function. Allows
704 * a group of CPUs to be quieted at one go, though all the CPUs in the
705 * group must be represented by the same leaf rcu_node structure.
706 * That structure's lock must be held upon entry, and it is released
710 cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
712 __releases(rnp->lock)
714 struct rcu_node *rnp_c;
716 /* Walk up the rcu_node hierarchy. */
718 if (!(rnp->qsmask & mask)) {
720 /* Our bit has already been cleared, so done. */
721 spin_unlock_irqrestore(&rnp->lock, flags);
724 rnp->qsmask &= ~mask;
725 if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) {
727 /* Other bits still set at this level, so done. */
728 spin_unlock_irqrestore(&rnp->lock, flags);
732 if (rnp->parent == NULL) {
734 /* No more levels. Exit loop holding root lock. */
738 spin_unlock_irqrestore(&rnp->lock, flags);
741 spin_lock_irqsave(&rnp->lock, flags);
742 WARN_ON_ONCE(rnp_c->qsmask);
746 * Get here if we are the last CPU to pass through a quiescent
747 * state for this grace period. Invoke cpu_quiet_msk_finish()
748 * to clean up and start the next grace period if one is needed.
750 cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */
754 * Record a quiescent state for the specified CPU, which must either be
755 * the current CPU. The lastcomp argument is used to make sure we are
756 * still in the grace period of interest. We don't want to end the current
757 * grace period based on quiescent states detected in an earlier grace
761 cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
765 struct rcu_node *rnp;
768 spin_lock_irqsave(&rnp->lock, flags);
769 if (lastcomp != ACCESS_ONCE(rsp->completed)) {
772 * Someone beat us to it for this grace period, so leave.
773 * The race with GP start is resolved by the fact that we
774 * hold the leaf rcu_node lock, so that the per-CPU bits
775 * cannot yet be initialized -- so we would simply find our
776 * CPU's bit already cleared in cpu_quiet_msk() if this race
779 rdp->passed_quiesc = 0; /* try again later! */
780 spin_unlock_irqrestore(&rnp->lock, flags);
784 if ((rnp->qsmask & mask) == 0) {
785 spin_unlock_irqrestore(&rnp->lock, flags);
790 * This GP can't end until cpu checks in, so all of our
791 * callbacks can be processed during the next GP.
793 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
795 cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */
800 * Check to see if there is a new grace period of which this CPU
801 * is not yet aware, and if so, set up local rcu_data state for it.
802 * Otherwise, see if this CPU has just passed through its first
803 * quiescent state for this grace period, and record that fact if so.
806 rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
808 /* If there is now a new grace period, record and return. */
809 if (check_for_new_grace_period(rsp, rdp))
813 * Does this CPU still need to do its part for current grace period?
814 * If no, return and let the other CPUs do their part as well.
816 if (!rdp->qs_pending)
820 * Was there a quiescent state since the beginning of the grace
821 * period? If no, then exit and wait for the next call.
823 if (!rdp->passed_quiesc)
826 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
827 cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
830 #ifdef CONFIG_HOTPLUG_CPU
833 * Move a dying CPU's RCU callbacks to the ->orphan_cbs_list for the
834 * specified flavor of RCU. The callbacks will be adopted by the next
835 * _rcu_barrier() invocation or by the CPU_DEAD notifier, whichever
836 * comes first. Because this is invoked from the CPU_DYING notifier,
837 * irqs are already disabled.
839 static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
842 struct rcu_data *rdp = rsp->rda[smp_processor_id()];
844 if (rdp->nxtlist == NULL)
845 return; /* irqs disabled, so comparison is stable. */
846 spin_lock(&rsp->onofflock); /* irqs already disabled. */
847 *rsp->orphan_cbs_tail = rdp->nxtlist;
848 rsp->orphan_cbs_tail = rdp->nxttail[RCU_NEXT_TAIL];
850 for (i = 0; i < RCU_NEXT_SIZE; i++)
851 rdp->nxttail[i] = &rdp->nxtlist;
852 rsp->orphan_qlen += rdp->qlen;
854 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
858 * Adopt previously orphaned RCU callbacks.
860 static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
863 struct rcu_data *rdp;
865 spin_lock_irqsave(&rsp->onofflock, flags);
866 rdp = rsp->rda[smp_processor_id()];
867 if (rsp->orphan_cbs_list == NULL) {
868 spin_unlock_irqrestore(&rsp->onofflock, flags);
871 *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_list;
872 rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_tail;
873 rdp->qlen += rsp->orphan_qlen;
874 rsp->orphan_cbs_list = NULL;
875 rsp->orphan_cbs_tail = &rsp->orphan_cbs_list;
876 rsp->orphan_qlen = 0;
877 spin_unlock_irqrestore(&rsp->onofflock, flags);
881 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
882 * and move all callbacks from the outgoing CPU to the current one.
884 static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
889 struct rcu_data *rdp = rsp->rda[cpu];
890 struct rcu_node *rnp;
892 /* Exclude any attempts to start a new grace period. */
893 spin_lock_irqsave(&rsp->onofflock, flags);
895 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
896 rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */
897 mask = rdp->grpmask; /* rnp->grplo is constant. */
899 spin_lock(&rnp->lock); /* irqs already disabled. */
900 rnp->qsmaskinit &= ~mask;
901 if (rnp->qsmaskinit != 0) {
902 spin_unlock(&rnp->lock); /* irqs remain disabled. */
907 * If there was a task blocking the current grace period,
908 * and if all CPUs have checked in, we need to propagate
909 * the quiescent state up the rcu_node hierarchy. But that
910 * is inconvenient at the moment due to deadlock issues if
911 * this should end the current grace period. So set the
912 * offlined CPU's bit in ->qsmask in order to force the
913 * next force_quiescent_state() invocation to clean up this
914 * mess in a deadlock-free manner.
916 if (rcu_preempt_offline_tasks(rsp, rnp, rdp) && !rnp->qsmask)
920 spin_unlock(&rnp->lock); /* irqs remain disabled. */
922 } while (rnp != NULL);
923 lastcomp = rsp->completed;
925 spin_unlock_irqrestore(&rsp->onofflock, flags);
927 rcu_adopt_orphan_cbs(rsp);
931 * Remove the specified CPU from the RCU hierarchy and move any pending
932 * callbacks that it might have to the current CPU. This code assumes
933 * that at least one CPU in the system will remain running at all times.
934 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
936 static void rcu_offline_cpu(int cpu)
938 __rcu_offline_cpu(cpu, &rcu_sched_state);
939 __rcu_offline_cpu(cpu, &rcu_bh_state);
940 rcu_preempt_offline_cpu(cpu);
943 #else /* #ifdef CONFIG_HOTPLUG_CPU */
945 static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
949 static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
953 static void rcu_offline_cpu(int cpu)
957 #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
960 * Invoke any RCU callbacks that have made it to the end of their grace
961 * period. Thottle as specified by rdp->blimit.
963 static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
966 struct rcu_head *next, *list, **tail;
969 /* If no callbacks are ready, just return.*/
970 if (!cpu_has_callbacks_ready_to_invoke(rdp))
974 * Extract the list of ready callbacks, disabling to prevent
975 * races with call_rcu() from interrupt handlers.
977 local_irq_save(flags);
979 rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
980 *rdp->nxttail[RCU_DONE_TAIL] = NULL;
981 tail = rdp->nxttail[RCU_DONE_TAIL];
982 for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
983 if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
984 rdp->nxttail[count] = &rdp->nxtlist;
985 local_irq_restore(flags);
987 /* Invoke callbacks. */
994 if (++count >= rdp->blimit)
998 local_irq_save(flags);
1000 /* Update count, and requeue any remaining callbacks. */
1003 *tail = rdp->nxtlist;
1004 rdp->nxtlist = list;
1005 for (count = 0; count < RCU_NEXT_SIZE; count++)
1006 if (&rdp->nxtlist == rdp->nxttail[count])
1007 rdp->nxttail[count] = tail;
1012 /* Reinstate batch limit if we have worked down the excess. */
1013 if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
1014 rdp->blimit = blimit;
1016 /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */
1017 if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) {
1018 rdp->qlen_last_fqs_check = 0;
1019 rdp->n_force_qs_snap = rsp->n_force_qs;
1020 } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark)
1021 rdp->qlen_last_fqs_check = rdp->qlen;
1023 local_irq_restore(flags);
1025 /* Re-raise the RCU softirq if there are callbacks remaining. */
1026 if (cpu_has_callbacks_ready_to_invoke(rdp))
1027 raise_softirq(RCU_SOFTIRQ);
1031 * Check to see if this CPU is in a non-context-switch quiescent state
1032 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
1033 * Also schedule the RCU softirq handler.
1035 * This function must be called with hardirqs disabled. It is normally
1036 * invoked from the scheduling-clock interrupt. If rcu_pending returns
1037 * false, there is no point in invoking rcu_check_callbacks().
1039 void rcu_check_callbacks(int cpu, int user)
1041 if (!rcu_pending(cpu))
1042 return; /* if nothing for RCU to do. */
1044 (idle_cpu(cpu) && rcu_scheduler_active &&
1045 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
1048 * Get here if this CPU took its interrupt from user
1049 * mode or from the idle loop, and if this is not a
1050 * nested interrupt. In this case, the CPU is in
1051 * a quiescent state, so note it.
1053 * No memory barrier is required here because both
1054 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1055 * variables that other CPUs neither access nor modify,
1056 * at least not while the corresponding CPU is online.
1062 } else if (!in_softirq()) {
1065 * Get here if this CPU did not take its interrupt from
1066 * softirq, in other words, if it is not interrupting
1067 * a rcu_bh read-side critical section. This is an _bh
1068 * critical section, so note it.
1073 rcu_preempt_check_callbacks(cpu);
1074 raise_softirq(RCU_SOFTIRQ);
1080 * Scan the leaf rcu_node structures, processing dyntick state for any that
1081 * have not yet encountered a quiescent state, using the function specified.
1082 * Returns 1 if the current grace period ends while scanning (possibly
1083 * because we made it end).
1085 static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp,
1086 int (*f)(struct rcu_data *))
1090 unsigned long flags;
1092 struct rcu_node *rnp;
1094 rcu_for_each_leaf_node(rsp, rnp) {
1096 spin_lock_irqsave(&rnp->lock, flags);
1097 if (rsp->completed != lastcomp) {
1098 spin_unlock_irqrestore(&rnp->lock, flags);
1101 if (rnp->qsmask == 0) {
1102 spin_unlock_irqrestore(&rnp->lock, flags);
1107 for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
1108 if ((rnp->qsmask & bit) != 0 && f(rsp->rda[cpu]))
1111 if (mask != 0 && rsp->completed == lastcomp) {
1113 /* cpu_quiet_msk() releases rnp->lock. */
1114 cpu_quiet_msk(mask, rsp, rnp, flags);
1117 spin_unlock_irqrestore(&rnp->lock, flags);
1123 * Force quiescent states on reluctant CPUs, and also detect which
1124 * CPUs are in dyntick-idle mode.
1126 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1128 unsigned long flags;
1130 struct rcu_node *rnp = rcu_get_root(rsp);
1134 if (!rcu_gp_in_progress(rsp))
1135 return; /* No grace period in progress, nothing to force. */
1136 if (!spin_trylock_irqsave(&rsp->fqslock, flags)) {
1137 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
1138 return; /* Someone else is already on the job. */
1141 (long)(rsp->jiffies_force_qs - jiffies) >= 0)
1142 goto unlock_ret; /* no emergency and done recently. */
1144 spin_lock(&rnp->lock);
1145 lastcomp = rsp->completed;
1146 signaled = rsp->signaled;
1147 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1148 if (lastcomp == rsp->gpnum) {
1149 rsp->n_force_qs_ngp++;
1150 spin_unlock(&rnp->lock);
1151 goto unlock_ret; /* no GP in progress, time updated. */
1153 spin_unlock(&rnp->lock);
1158 break; /* grace period idle or initializing, ignore. */
1160 case RCU_SAVE_DYNTICK:
1162 if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK)
1163 break; /* So gcc recognizes the dead code. */
1165 /* Record dyntick-idle state. */
1166 if (rcu_process_dyntick(rsp, lastcomp,
1167 dyntick_save_progress_counter))
1169 /* fall into next case. */
1171 case RCU_SAVE_COMPLETED:
1173 /* Update state, record completion counter. */
1175 spin_lock(&rnp->lock);
1176 if (lastcomp == rsp->completed &&
1177 rsp->signaled == signaled) {
1178 rsp->signaled = RCU_FORCE_QS;
1179 dyntick_record_completed(rsp, lastcomp);
1180 forcenow = signaled == RCU_SAVE_COMPLETED;
1182 spin_unlock(&rnp->lock);
1185 /* fall into next case. */
1189 /* Check dyntick-idle state, send IPI to laggarts. */
1190 if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp),
1191 rcu_implicit_dynticks_qs))
1194 /* Leave state in case more forcing is required. */
1199 spin_unlock_irqrestore(&rsp->fqslock, flags);
1202 #else /* #ifdef CONFIG_SMP */
1204 static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1209 #endif /* #else #ifdef CONFIG_SMP */
1212 * This does the RCU processing work from softirq context for the
1213 * specified rcu_state and rcu_data structures. This may be called
1214 * only from the CPU to whom the rdp belongs.
1217 __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1219 unsigned long flags;
1221 WARN_ON_ONCE(rdp->beenonline == 0);
1224 * If an RCU GP has gone long enough, go check for dyntick
1225 * idle CPUs and, if needed, send resched IPIs.
1227 if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1228 force_quiescent_state(rsp, 1);
1231 * Advance callbacks in response to end of earlier grace
1232 * period that some other CPU ended.
1234 rcu_process_gp_end(rsp, rdp);
1236 /* Update RCU state based on any recent quiescent states. */
1237 rcu_check_quiescent_state(rsp, rdp);
1239 /* Does this CPU require a not-yet-started grace period? */
1240 if (cpu_needs_another_gp(rsp, rdp)) {
1241 spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags);
1242 rcu_start_gp(rsp, flags); /* releases above lock */
1245 /* If there are callbacks ready, invoke them. */
1246 rcu_do_batch(rsp, rdp);
1250 * Do softirq processing for the current CPU.
1252 static void rcu_process_callbacks(struct softirq_action *unused)
1255 * Memory references from any prior RCU read-side critical sections
1256 * executed by the interrupted code must be seen before any RCU
1257 * grace-period manipulations below.
1259 smp_mb(); /* See above block comment. */
1261 __rcu_process_callbacks(&rcu_sched_state,
1262 &__get_cpu_var(rcu_sched_data));
1263 __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1264 rcu_preempt_process_callbacks();
1267 * Memory references from any later RCU read-side critical sections
1268 * executed by the interrupted code must be seen after any RCU
1269 * grace-period manipulations above.
1271 smp_mb(); /* See above block comment. */
1275 __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1276 struct rcu_state *rsp)
1278 unsigned long flags;
1279 struct rcu_data *rdp;
1284 smp_mb(); /* Ensure RCU update seen before callback registry. */
1287 * Opportunistically note grace-period endings and beginnings.
1288 * Note that we might see a beginning right after we see an
1289 * end, but never vice versa, since this CPU has to pass through
1290 * a quiescent state betweentimes.
1292 local_irq_save(flags);
1293 rdp = rsp->rda[smp_processor_id()];
1294 rcu_process_gp_end(rsp, rdp);
1295 check_for_new_grace_period(rsp, rdp);
1297 /* Add the callback to our list. */
1298 *rdp->nxttail[RCU_NEXT_TAIL] = head;
1299 rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1301 /* Start a new grace period if one not already started. */
1302 if (!rcu_gp_in_progress(rsp)) {
1303 unsigned long nestflag;
1304 struct rcu_node *rnp_root = rcu_get_root(rsp);
1306 spin_lock_irqsave(&rnp_root->lock, nestflag);
1307 rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */
1311 * Force the grace period if too many callbacks or too long waiting.
1312 * Enforce hysteresis, and don't invoke force_quiescent_state()
1313 * if some other CPU has recently done so. Also, don't bother
1314 * invoking force_quiescent_state() if the newly enqueued callback
1315 * is the only one waiting for a grace period to complete.
1317 if (unlikely(++rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) {
1318 rdp->blimit = LONG_MAX;
1319 if (rsp->n_force_qs == rdp->n_force_qs_snap &&
1320 *rdp->nxttail[RCU_DONE_TAIL] != head)
1321 force_quiescent_state(rsp, 0);
1322 rdp->n_force_qs_snap = rsp->n_force_qs;
1323 rdp->qlen_last_fqs_check = rdp->qlen;
1324 } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)
1325 force_quiescent_state(rsp, 1);
1326 local_irq_restore(flags);
1330 * Queue an RCU-sched callback for invocation after a grace period.
1332 void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1334 __call_rcu(head, func, &rcu_sched_state);
1336 EXPORT_SYMBOL_GPL(call_rcu_sched);
1339 * Queue an RCU for invocation after a quicker grace period.
1341 void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1343 __call_rcu(head, func, &rcu_bh_state);
1345 EXPORT_SYMBOL_GPL(call_rcu_bh);
1348 * Check to see if there is any immediate RCU-related work to be done
1349 * by the current CPU, for the specified type of RCU, returning 1 if so.
1350 * The checks are in order of increasing expense: checks that can be
1351 * carried out against CPU-local state are performed first. However,
1352 * we must check for CPU stalls first, else we might not get a chance.
1354 static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1356 rdp->n_rcu_pending++;
1358 /* Check for CPU stalls, if enabled. */
1359 check_cpu_stall(rsp, rdp);
1361 /* Is the RCU core waiting for a quiescent state from this CPU? */
1362 if (rdp->qs_pending) {
1363 rdp->n_rp_qs_pending++;
1367 /* Does this CPU have callbacks ready to invoke? */
1368 if (cpu_has_callbacks_ready_to_invoke(rdp)) {
1369 rdp->n_rp_cb_ready++;
1373 /* Has RCU gone idle with this CPU needing another grace period? */
1374 if (cpu_needs_another_gp(rsp, rdp)) {
1375 rdp->n_rp_cpu_needs_gp++;
1379 /* Has another RCU grace period completed? */
1380 if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */
1381 rdp->n_rp_gp_completed++;
1385 /* Has a new RCU grace period started? */
1386 if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */
1387 rdp->n_rp_gp_started++;
1391 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1392 if (rcu_gp_in_progress(rsp) &&
1393 ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) {
1394 rdp->n_rp_need_fqs++;
1399 rdp->n_rp_need_nothing++;
1404 * Check to see if there is any immediate RCU-related work to be done
1405 * by the current CPU, returning 1 if so. This function is part of the
1406 * RCU implementation; it is -not- an exported member of the RCU API.
1408 static int rcu_pending(int cpu)
1410 return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) ||
1411 __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) ||
1412 rcu_preempt_pending(cpu);
1416 * Check to see if any future RCU-related work will need to be done
1417 * by the current CPU, even if none need be done immediately, returning
1418 * 1 if so. This function is part of the RCU implementation; it is -not-
1419 * an exported member of the RCU API.
1421 int rcu_needs_cpu(int cpu)
1423 /* RCU callbacks either ready or pending? */
1424 return per_cpu(rcu_sched_data, cpu).nxtlist ||
1425 per_cpu(rcu_bh_data, cpu).nxtlist ||
1426 rcu_preempt_needs_cpu(cpu);
1429 static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL};
1430 static atomic_t rcu_barrier_cpu_count;
1431 static DEFINE_MUTEX(rcu_barrier_mutex);
1432 static struct completion rcu_barrier_completion;
1434 static void rcu_barrier_callback(struct rcu_head *notused)
1436 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
1437 complete(&rcu_barrier_completion);
1441 * Called with preemption disabled, and from cross-cpu IRQ context.
1443 static void rcu_barrier_func(void *type)
1445 int cpu = smp_processor_id();
1446 struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
1447 void (*call_rcu_func)(struct rcu_head *head,
1448 void (*func)(struct rcu_head *head));
1450 atomic_inc(&rcu_barrier_cpu_count);
1451 call_rcu_func = type;
1452 call_rcu_func(head, rcu_barrier_callback);
1456 * Orchestrate the specified type of RCU barrier, waiting for all
1457 * RCU callbacks of the specified type to complete.
1459 static void _rcu_barrier(struct rcu_state *rsp,
1460 void (*call_rcu_func)(struct rcu_head *head,
1461 void (*func)(struct rcu_head *head)))
1463 BUG_ON(in_interrupt());
1464 /* Take mutex to serialize concurrent rcu_barrier() requests. */
1465 mutex_lock(&rcu_barrier_mutex);
1466 init_completion(&rcu_barrier_completion);
1468 * Initialize rcu_barrier_cpu_count to 1, then invoke
1469 * rcu_barrier_func() on each CPU, so that each CPU also has
1470 * incremented rcu_barrier_cpu_count. Only then is it safe to
1471 * decrement rcu_barrier_cpu_count -- otherwise the first CPU
1472 * might complete its grace period before all of the other CPUs
1473 * did their increment, causing this function to return too
1476 atomic_set(&rcu_barrier_cpu_count, 1);
1477 preempt_disable(); /* stop CPU_DYING from filling orphan_cbs_list */
1478 rcu_adopt_orphan_cbs(rsp);
1479 on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1);
1480 preempt_enable(); /* CPU_DYING can again fill orphan_cbs_list */
1481 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
1482 complete(&rcu_barrier_completion);
1483 wait_for_completion(&rcu_barrier_completion);
1484 mutex_unlock(&rcu_barrier_mutex);
1488 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
1490 void rcu_barrier_bh(void)
1492 _rcu_barrier(&rcu_bh_state, call_rcu_bh);
1494 EXPORT_SYMBOL_GPL(rcu_barrier_bh);
1497 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
1499 void rcu_barrier_sched(void)
1501 _rcu_barrier(&rcu_sched_state, call_rcu_sched);
1503 EXPORT_SYMBOL_GPL(rcu_barrier_sched);
1506 * Do boot-time initialization of a CPU's per-CPU RCU data.
1509 rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
1511 unsigned long flags;
1513 struct rcu_data *rdp = rsp->rda[cpu];
1514 struct rcu_node *rnp = rcu_get_root(rsp);
1516 /* Set up local state, ensuring consistent view of global state. */
1517 spin_lock_irqsave(&rnp->lock, flags);
1518 rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1519 rdp->nxtlist = NULL;
1520 for (i = 0; i < RCU_NEXT_SIZE; i++)
1521 rdp->nxttail[i] = &rdp->nxtlist;
1524 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1525 #endif /* #ifdef CONFIG_NO_HZ */
1527 spin_unlock_irqrestore(&rnp->lock, flags);
1531 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1532 * offline event can be happening at a given time. Note also that we
1533 * can accept some slop in the rsp->completed access due to the fact
1534 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1536 static void __cpuinit
1537 rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable)
1539 unsigned long flags;
1542 struct rcu_data *rdp = rsp->rda[cpu];
1543 struct rcu_node *rnp = rcu_get_root(rsp);
1545 /* Set up local state, ensuring consistent view of global state. */
1546 spin_lock_irqsave(&rnp->lock, flags);
1547 lastcomp = rsp->completed;
1548 rdp->completed = lastcomp;
1549 rdp->gpnum = lastcomp;
1550 rdp->passed_quiesc = 0; /* We could be racing with new GP, */
1551 rdp->qs_pending = 1; /* so set up to respond to current GP. */
1552 rdp->beenonline = 1; /* We have now been online. */
1553 rdp->preemptable = preemptable;
1554 rdp->passed_quiesc_completed = lastcomp - 1;
1555 rdp->qlen_last_fqs_check = 0;
1556 rdp->n_force_qs_snap = rsp->n_force_qs;
1557 rdp->blimit = blimit;
1558 spin_unlock(&rnp->lock); /* irqs remain disabled. */
1561 * A new grace period might start here. If so, we won't be part
1562 * of it, but that is OK, as we are currently in a quiescent state.
1565 /* Exclude any attempts to start a new GP on large systems. */
1566 spin_lock(&rsp->onofflock); /* irqs already disabled. */
1568 /* Add CPU to rcu_node bitmasks. */
1570 mask = rdp->grpmask;
1572 /* Exclude any attempts to start a new GP on small systems. */
1573 spin_lock(&rnp->lock); /* irqs already disabled. */
1574 rnp->qsmaskinit |= mask;
1575 mask = rnp->grpmask;
1576 spin_unlock(&rnp->lock); /* irqs already disabled. */
1578 } while (rnp != NULL && !(rnp->qsmaskinit & mask));
1580 spin_unlock_irqrestore(&rsp->onofflock, flags);
1583 static void __cpuinit rcu_online_cpu(int cpu)
1585 rcu_init_percpu_data(cpu, &rcu_sched_state, 0);
1586 rcu_init_percpu_data(cpu, &rcu_bh_state, 0);
1587 rcu_preempt_init_percpu_data(cpu);
1591 * Handle CPU online/offline notification events.
1593 int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1594 unsigned long action, void *hcpu)
1596 long cpu = (long)hcpu;
1599 case CPU_UP_PREPARE:
1600 case CPU_UP_PREPARE_FROZEN:
1601 rcu_online_cpu(cpu);
1604 case CPU_DYING_FROZEN:
1606 * preempt_disable() in _rcu_barrier() prevents stop_machine(),
1607 * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);"
1608 * returns, all online cpus have queued rcu_barrier_func().
1609 * The dying CPU clears its cpu_online_mask bit and
1610 * moves all of its RCU callbacks to ->orphan_cbs_list
1611 * in the context of stop_machine(), so subsequent calls
1612 * to _rcu_barrier() will adopt these callbacks and only
1613 * then queue rcu_barrier_func() on all remaining CPUs.
1615 rcu_send_cbs_to_orphanage(&rcu_bh_state);
1616 rcu_send_cbs_to_orphanage(&rcu_sched_state);
1617 rcu_preempt_send_cbs_to_orphanage();
1620 case CPU_DEAD_FROZEN:
1621 case CPU_UP_CANCELED:
1622 case CPU_UP_CANCELED_FROZEN:
1623 rcu_offline_cpu(cpu);
1632 * Compute the per-level fanout, either using the exact fanout specified
1633 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1635 #ifdef CONFIG_RCU_FANOUT_EXACT
1636 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1640 for (i = NUM_RCU_LVLS - 1; i >= 0; i--)
1641 rsp->levelspread[i] = CONFIG_RCU_FANOUT;
1643 #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1644 static void __init rcu_init_levelspread(struct rcu_state *rsp)
1651 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1652 ccur = rsp->levelcnt[i];
1653 rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
1657 #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1660 * Helper function for rcu_init() that initializes one rcu_state structure.
1662 static void __init rcu_init_one(struct rcu_state *rsp)
1667 struct rcu_node *rnp;
1669 /* Initialize the level-tracking arrays. */
1671 for (i = 1; i < NUM_RCU_LVLS; i++)
1672 rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
1673 rcu_init_levelspread(rsp);
1675 /* Initialize the elements themselves, starting from the leaves. */
1677 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1678 cpustride *= rsp->levelspread[i];
1679 rnp = rsp->level[i];
1680 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
1681 if (rnp != rcu_get_root(rsp))
1682 spin_lock_init(&rnp->lock);
1685 rnp->qsmaskinit = 0;
1686 rnp->grplo = j * cpustride;
1687 rnp->grphi = (j + 1) * cpustride - 1;
1688 if (rnp->grphi >= NR_CPUS)
1689 rnp->grphi = NR_CPUS - 1;
1695 rnp->grpnum = j % rsp->levelspread[i - 1];
1696 rnp->grpmask = 1UL << rnp->grpnum;
1697 rnp->parent = rsp->level[i - 1] +
1698 j / rsp->levelspread[i - 1];
1701 INIT_LIST_HEAD(&rnp->blocked_tasks[0]);
1702 INIT_LIST_HEAD(&rnp->blocked_tasks[1]);
1705 spin_lock_init(&rcu_get_root(rsp)->lock);
1709 * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used
1710 * nowhere else! Assigns leaf node pointers into each CPU's rcu_data
1713 #define RCU_INIT_FLAVOR(rsp, rcu_data) \
1717 struct rcu_node *rnp; \
1719 rcu_init_one(rsp); \
1720 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1722 for_each_possible_cpu(i) { \
1723 if (i > rnp[j].grphi) \
1725 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1726 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1727 rcu_boot_init_percpu_data(i, rsp); \
1731 void __init __rcu_init(void)
1733 rcu_bootup_announce();
1734 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1735 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
1736 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1737 RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data);
1738 RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data);
1739 __rcu_init_preempt();
1740 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1743 #include "rcutree_plugin.h"